Re: [PATCH v2 3/7] of: resolver: Add export_symbols in of_resolve_phandles() parameters

Hello Hervé,

On Wed, 30 Apr 2025 14:51:47 +0200
Herve Codina <[email protected]> wrote:

> In order to prepare the introduction of the export symbols node
> handling, add a export_symbols parameter in of_resolve_phandles().
> 
> The export_symbols is the export symbols device tree node the resolver
> will use for the overlay symbols resolution.
> 
> Signed-off-by: Herve Codina <[email protected]>
> Tested-by: Ayush Singh <[email protected]>

[...]

> --- a/drivers/of/resolver.c
> +++ b/drivers/of/resolver.c
> @@ -237,7 +237,8 @@ static int get_phandle_from_symbols_node(const struct device_node *tree_symbols,
>  /**
>   * of_resolve_phandles - Relocate and resolve overlay against live tree
>   *
> - * @overlay:	Pointer to devicetree overlay to relocate and resolve
> + * @overlay:		Pointer to devicetree overlay to relocate and resolve
> + * @export_symbols:	Pointer to devicetree export symbols node.
>   *
>   * Modify (relocate) values of local phandles in @overlay to a range that
>   * does not conflict with the live expanded devicetree.  Update references
> @@ -257,6 +258,10 @@ static int get_phandle_from_symbols_node(const struct device_node *tree_symbols,
>   * corresponding to that symbol in the live tree.  Update the references in
>   * the overlay with the phandle values in the live tree.
>   *
> + * @export_symbols can be use in this references update. The resolver tries
> + * first to find a match in the @export_symbols. If not found, it uses the
> + * "__symbol__" node in the live tree.

The rationale behind this logic is not clear to me. I'd have expected
instead this logic:

  if (export-symbols != NULL):
      match only in export-symbols
  else
      match only in __symbols__

following the idea that it's better to be strict when introducing
something, and possibly relax it later on.

As I see it, with the current logic if you use export-symbols but you
build dtbs with -@, you can still match a global label. export-symbols
should avoid that instead.

Let me know whether I'm missing something here (which is surely
possible).

Other than this, the series looks good to me. I'm not adding my R-by on
the implementation patches though, waiting for the above question to be
clarified.

Luca

-- 
Luca Ceresoli, Bootlin
Embedded Linux and Kernel engineering
https://bootlin.com